• December 10, 2024
Maxim Nets $3 Million Funding Round to Provide Standardized Evaluation in AI Development

The race to build AI-based products—including automation—has resulted in a very different development cycle than the one tech companies have become accustomed to over decades of software development. A Mountain View, Calif.-based startup has raised $3 million to help bridge the gap between the way AI products are being built and the developers who are used to a very different environment.

Maxim, founded by Google-Postman vets Vaibhavi Gangwar and Akshay Deo, was launched because the development of applications driven by LLMs lack the “deterministic paradigm” marked by standardized best practices that traditional software had. AI-based products, the company says, is not deterministic, resulting in unpredictable variability based on the model, data, context or prompt.

Maxim’s platform sits between the foundational model and application layers of the AI stack, enabling end-to-end AI evaluation across the development lifecycle for AI developers who comprise the traditional AI/ML and backend engineers who are leading AI development.

“Even though we are seeing tremendous value creation in the AI hardware and foundational model stack, for the benefits to be realized meaningfully up the value chain, we need the products built on top of the generative AI stack to be trustworthy and reliable,” said Vaibhavi Gangwar and Akshay Deo, co-founders of Maxim. “This makes evaluation a very critical piece of AI infrastructure. However, today there is no standardization in the generative AI testing space unlike traditional software development. That’s where Maxim comes in: with our users, we have become this very core piece of infrastructure that is powering high quality AI development, streamlining team collaboration, and saving development teams hundreds of hours per month.”

The round was led by Elevation Capital with participation from angel investors including founders of Postman, Chargebee, Groww, Razorpay, and Media.net.